Add configurable MTP and shared-weight draft memory fitting
This commit is contained in:
@@ -26,8 +26,8 @@ class ManagementTests(unittest.TestCase):
|
||||
req.update(id=saved['id'],revision=1)
|
||||
self.profiles.save(req)
|
||||
restarted=Profiles(self.root/'profiles.json',self.catalog)
|
||||
self.assertEqual(restarted.status()['profiles'][0]['parameters'],params)
|
||||
for change in [dict(gpu_devices=devices*2),dict(tensor_split=[85]),dict(split_mode='none'),dict(top_p=1.1),dict(temperature=float('nan')),dict(top_k=True),dict(tensor_split=[0,100]),dict(gpu_devices=['CUDA0','CUDA1'])]:
|
||||
self.assertEqual(restarted.status()['profiles'][0]['parameters'],dict(params,mtp=False,mtp_tokens=2,mtp_min_p=.05))
|
||||
for change in [dict(mtp='yes'),dict(mtp_tokens=0),dict(mtp_tokens=True),dict(mtp_min_p=1.1),dict(gpu_devices=devices*2),dict(tensor_split=[85]),dict(split_mode='none'),dict(top_p=1.1),dict(temperature=float('nan')),dict(top_k=True),dict(tensor_split=[0,100]),dict(gpu_devices=['CUDA0','CUDA1'])]:
|
||||
with self.subTest(change=change),self.assertRaises(ValueError):self.profiles.save(dict(req,revision=2,parameters=dict(params,**change)))
|
||||
|
||||
def test_profiles_persist_and_update_conflicts(self):
|
||||
|
||||
Reference in New Issue
Block a user